Development and assessment of automated forest road projection methods using performance metrics relevant for wildlife

Abstract

Context: Resource road networks have complex and varied impacts on wildlife and other forest values, yet spatial stochastic models forecasting impacts of forest disturbance rarely include automated road network projections. Hardy et al. (Can J For Res 2023. 10.1139/cjfr-2022-0306) partially addressed this need with a LANDIS-II extension, but there remains a need for tools that can be integrated into other modelling frameworks while identifying a pragmatic balance between achieving ecological relevancy and computational cost. Objectives: Our goal is an open source resource road network projection tool that can be easily incorporated into modelling frameworks that assess the implications of forest change for wildlife. We compared the performance of several resource road network projection methods using ecologically relevant metrics. Methods: We implemented simple iterative least cost path and minimum spanning tree methods with grade penalties in the open source R roads package. We assessed performance by comparing projections to observed resource road development since 1990 in a mountainous region of British Columbia. Results: All resource road projection methods that we tested performed relatively well. Grade penalties improved performance, as did our minimum spanning tree method. However, the minimum spanning tree method required more computing time and memory, so users must weigh the benefits of improved performance against computational costs. Conclusions: Our resource road network simulation methods can improve projections of anticipated resource development impacts on wildlife across large areas. Our open source implementation in the R roads package will be useful for improving projections of the cumulative effects of natural and anthropogenic disturbances on wildlife in an era of rapid change.